Research map / Architectures and efficiency / RNNs and normalization
Efficient Beam Tree Recursion
Jishnu Ray Chowdhury, Cornelia Caragea
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add slicing of hidden state for scorer (ablation) | EBT-GRC · ListOps seq len 900-1000 | peak memory (GB): 5.4 -> 2.78 | Improved |
| add slicing of hidden state for scorer (ablation) | EBT-GRC · ListOps seq len 1500-2000 | peak memory (GB): 21.12 -> 10.97 | Improved |
| add slicing with hidden size 512 (ablation) | EBT-GRC (512) · ListOps seq len 1500-2000 | peak memory (GB): OOM -> 11.41 | Improved |
| add slicing of hidden state for scorer (ablation) | EBT-GRC · ListOps LRA 2000 | accuracy: 75.5 -> 79.3 | Improved |
| add slicing of hidden state for scorer (ablation) | EBT-GRC · ListOps length gen 900-1000 | accuracy: 98.6 -> 99.3 | Improved |
| use efficient EBT-GRC instead of BT-GRC OS | BT-RvNN · ListOps seq len 900-1000 | peak memory (GB): 42.95 -> 2.78 | Improved |
| use efficient EBT-GRC instead of BT-GRC | BT-RvNN · ListOps length gen 900-1000 | accuracy: 88.4 -> 99.3 | Improved |
| use EBT-GRC instead of BT-GRC | BT-RvNN · ListOps seq len 900-1000 | time (min): 5.1 -> 5.5 | No clear effect |
| use EBT-GRC instead of OM | Ordered Memory baseline · ListOps seq len 900-1000 | time (min): 38.2 -> 5.5 | Improved |
| use EBT-GRC instead of BT-GRC | BT-RvNN · SNLI IID | accuracy: 84.9 -> 84.7 | No clear effect |
| use beam tree EBT-GAU instead of plain GAU | sequence interaction model · SNLI IID | accuracy: 87 -> 87.5 | Improved |
| use beam search EBT-GAU instead of greedy EGT-GAU | sequence interaction model · SNLI Hard | accuracy: 74.8 -> 75.7 | Improved |
Closest papers by meaning
- Beam Tree Recursive Cells (ICML 2023 · similarity 0.92)
- Recursion in Recursion: Two-Level Nested Recursion for Length Generalization with Scalability (NeurIPS 2023 · similarity 0.90)
- An Efficient Matrix Multiplication Algorithm for Accelerating Inference in Binary and Ternary Neural Networks (ICML 2025 · similarity 0.82)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · similarity 0.82)
- Separations in the Representational Capabilities of Transformers and Recurrent Architectures (NeurIPS 2024 · similarity 0.82)
- Rethinking the Capacity of Graph Neural Networks for Branching Strategy (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
